Re: Fixes for uaccess.h with gcc >= 4.0.1

[Thiemo Seufer <ths@networkno.de>]

On Mon, Jan 23, 2006 at 05:50:53PM +0200, P. Christeas wrote:
> On Monday 23 January 2006 5:37 pm, Ralf Baechle wrote:
> > On Mon, Jan 23, 2006 at 05:18:38PM +0200, P. Christeas wrote:
> > > On Monday 23 January 2006 5:05 pm, Ralf Baechle wrote:
> > > > I'd appreciate if somebody with gcc 4.0.1 could test this kernel patch
> > > > below.
> > > >
> > > >   Ralf
> > >
> > > Is that for 2.4?
> >
> > 2.4 is a no go for all architectures with gcc >= 4.0.0 and in case of MIPS
> > even gcc 3.4 is somewhat dubious.
> >
> > > 2.6 doesn't seem to have that problem..
> >
> > It's probably a matter of configuration then.  Basically with our current
> > uaccess.h and gcc >= 4.0.1 the attempt to pass a pointer to a const
> > variable as the pointer argument to get_user or __get_user will blow up.
> > It's always been a bug - but gcc before 4.0.1 were accepting this
> > silently.
> >
> >   Ralf
> 
> I 've been compiling with gcc 4.0.2 (my tree is Linus') and haven't seen any 
> message like that.

The case I saw happened for 32bit compat ioctls in a 64bit kernel.

> It all compiles fine. Is there a point in testing your 
> patch as well?

Well, if you want to be sure it doesn't break your system...


Thiemo